AFCL: Analytic Federated Continual Learning for Spatio-Temporal Invariance of Non-IID Data
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916742274482176 |
|---|---|
| author | Tang, Jianheng Zhuang, Huiping He, Jingyu He, Run Wang, Jingchao Fan, Kejia Liu, Anfeng Wang, Tian Wang, Leye Zhu, Zhanxing Zhang, Shanghang Song, Houbing Herbert Liu, Yunhuai |
| author_facet | Tang, Jianheng Zhuang, Huiping He, Jingyu He, Run Wang, Jingchao Fan, Kejia Liu, Anfeng Wang, Tian Wang, Leye Zhu, Zhanxing Zhang, Shanghang Song, Houbing Herbert Liu, Yunhuai |
| contents | Federated Continual Learning (FCL) enables distributed clients to collaboratively train a global model from online task streams in dynamic real-world scenarios. However, existing FCL methods face challenges of both spatial data heterogeneity among distributed clients and temporal data heterogeneity across online tasks. Such data heterogeneity significantly degrades the model performance with severe spatial-temporal catastrophic forgetting of local and past knowledge. In this paper, we identify that the root cause of this issue lies in the inherent vulnerability and sensitivity of gradients to non-IID data. To fundamentally address this issue, we propose a gradient-free method, named Analytic Federated Continual Learning (AFCL), by deriving analytical (i.e., closed-form) solutions from frozen extracted features. In local training, our AFCL enables single-epoch learning with only a lightweight forward-propagation process for each client. In global aggregation, the server can recursively and efficiently update the global model with single-round aggregation. Theoretical analyses validate that our AFCL achieves spatio-temporal invariance of non-IID data. This ideal property implies that, regardless of how heterogeneous the data are distributed across local clients and online tasks, the aggregated model of our AFCL remains invariant and identical to that of centralized joint learning. Extensive experiments show the consistent superiority of our AFCL over state-of-the-art baselines across various benchmark datasets and settings. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_12245 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | AFCL: Analytic Federated Continual Learning for Spatio-Temporal Invariance of Non-IID Data Tang, Jianheng Zhuang, Huiping He, Jingyu He, Run Wang, Jingchao Fan, Kejia Liu, Anfeng Wang, Tian Wang, Leye Zhu, Zhanxing Zhang, Shanghang Song, Houbing Herbert Liu, Yunhuai Machine Learning Artificial Intelligence Federated Continual Learning (FCL) enables distributed clients to collaboratively train a global model from online task streams in dynamic real-world scenarios. However, existing FCL methods face challenges of both spatial data heterogeneity among distributed clients and temporal data heterogeneity across online tasks. Such data heterogeneity significantly degrades the model performance with severe spatial-temporal catastrophic forgetting of local and past knowledge. In this paper, we identify that the root cause of this issue lies in the inherent vulnerability and sensitivity of gradients to non-IID data. To fundamentally address this issue, we propose a gradient-free method, named Analytic Federated Continual Learning (AFCL), by deriving analytical (i.e., closed-form) solutions from frozen extracted features. In local training, our AFCL enables single-epoch learning with only a lightweight forward-propagation process for each client. In global aggregation, the server can recursively and efficiently update the global model with single-round aggregation. Theoretical analyses validate that our AFCL achieves spatio-temporal invariance of non-IID data. This ideal property implies that, regardless of how heterogeneous the data are distributed across local clients and online tasks, the aggregated model of our AFCL remains invariant and identical to that of centralized joint learning. Extensive experiments show the consistent superiority of our AFCL over state-of-the-art baselines across various benchmark datasets and settings. |
| title | AFCL: Analytic Federated Continual Learning for Spatio-Temporal Invariance of Non-IID Data |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2505.12245 |